Towards Generalizable Protein-ligand Co-folding with ACER
- 1Eindhoven University of Technology
Abstract
Predicting protein-ligand complex structures is a central challenge in drug discovery. While recent co-folding models such as AlphaFold-3 achieve accurate structure prediction, they fail to generalize to underexplored binding interfaces - systematically misplacing ligands, particularly for allosteric or structurally novel targets. To address this gap, we present ACER (A daptive Co-folding via pocket E xploration and pose R anking), a training-free framework that (a) enables co-folding models to systematically explore alternative binding pockets, and (b) leverages the discovered pockets to increase pose accuracy. Our method enables the efficient discovery of non-prevalent pockets without prior expert knowledge. ACER improves pocket discovery and pose accuracy on allosteric targets and structurally novel complexes, successfully modeling binding interfaces that are under-represented or absent from the training set. Our results demonstrate how improved sampling dynamics enhance the generalisability of co-folding models without retraining.
Related
- Targeted finetuning enables co-folding models to learn ligand-induced protein conformational states
- FoldAbS: Repurposing the Protein Folding Model as a Foundation Encoder for Antibody Screening
- PathFold: Predicting the Entire Protein Folding Pathway from Protein Sequence Alone
- RPFace: From biased structural benchmarks to generalizable RNA-protein interface prediction
- Targeting the Ero1α-PDI oxidative protein folding pathway as a broad-spectrum strategy against coronavirus.
BibTeX
@misc{vithayapalert2026towards,
title = {{Towards Generalizable Protein-ligand Co-folding with ACER}},
author = {N. Vithayapalert and F. Grisoni},
howpublished = {bioRxiv},
year = {2026},
month = jun,
doi = {10.64898/2026.06.02.728568},
url = {https://www.biorxiv.org/content/10.64898/2026.06.02.728568v2},
note = {CC BY},
}
RIS
TY - UNPB
TI - Towards Generalizable Protein-ligand Co-folding with ACER
AU - N. Vithayapalert
AU - F. Grisoni
PY - 2026
DA - 2026/06/02
PB - bioRxiv
DO - 10.64898/2026.06.02.728568
UR - https://www.biorxiv.org/content/10.64898/2026.06.02.728568v2
AB - Predicting protein-ligand complex structures is a central challenge in drug discovery. While recent co-folding models such as AlphaFold-3 achieve accurate structure prediction, they fail to generalize to underexplored binding interfaces - systematically misplacing ligands, particularly for allosteric or structurally novel targets. To address this gap, we present ACER (A daptive Co-folding via pocket E xploration and pose R anking), a training-free framework that (a) enables co-folding models to systematically explore alternative binding pockets, and (b) leverages the discovered pockets to increase pose accuracy. Our method enables the efficient discovery of non-prevalent pockets without prior expert knowledge. ACER improves pocket discovery and pose accuracy on allosteric targets and structurally novel complexes, successfully modeling binding interfaces that are under-represented or absent from the training set. Our results demonstrate how improved sampling dynamics enhance the generalisability of co-folding models without retraining.
ER -